Anthropic is expanding its Cyber Verification Program to allow a broader range of security professionals access to Claude models with reduced safety filters. This change permits vetted individuals and organisations to use the AI for vulnerability research, malware analysis, incident response, and penetration testing. Previously, public versions of the tool blocked most of these activities because similar capabilities could assist attackers. Basic tasks such as code review remain available without special permission.
The updated structure divides access into three distinct tiers. Defense Access covers corporate teams, government bodies, universities, and individual researchers. Red Team Access restricts authorised attack simulations to organisations only. Specialized Access is reserved for testing critical systems like power grids or flight controls and requires joint vetting with the US government. Partners in the predecessor program, Project Glasswing, identified at least 129,000 confirmed vulnerabilities between April and July 2026. More than 33,000 of these were classified as high-severity or critical. Several partners reported that the AI accelerated their work by months or years.
* The reported vulnerability numbers come from surveys of a subset of partners.
* Anthropic estimates the real-world impact is at least five times higher than reported.
* Individual researchers can now join the Defense Access tier.
